Nemotron-RL-Instruction-Following-MultiTurnChat-v1-prompt-only
收藏资源简介:
该数据集名为Nemotron-RL-Instruction-Following-MultiTurnChat-v1-prompt-only,是从源数据集nvidia/Nemotron-RL-Instruction-Following-MultiTurnChat-v1中提取的仅包含提示(prompt-only)的版本。数据内容主要包括prompts.csv文件,其中每条记录对应源数据集的一行,包含提取的prompt、可选的独立system_prompt以及当源行定义可用工具时的结构化tools字段,嵌套值在CSV单元格中以JSON格式编码。数据集还包含两个辅助文件:summary.md提供源行数、提取行数、计数差异和失败提示计数的摘要;null_or_empty_rows.md列出提示提取结果为空或null的行索引。数据规模为提取行数2011行,失败提示行数为0,行数差异为0。该数据集适用于需要多轮对话指令跟随的提示工程、强化学习后训练或语言模型微调任务,尤其关注提示的提取和结构化表示。数据集由Nemotron后训练v3提示提取工作流生成,并通过jamesdborin上传。
The dataset is named Nemotron-RL-Instruction-Following-MultiTurnChat-v1-prompt-only, extracted from the source dataset nvidia/Nemotron-RL-Instruction-Following-MultiTurnChat-v1 as a prompt-only version. The data primarily consists of a prompts.csv file, where each record corresponds to a row in the source dataset and includes the extracted prompt, an optional independent system_prompt, and a structured tools field when tools are defined in the source row, with nested values encoded in JSON format within CSV cells. The dataset also includes two auxiliary files: summary.md provides a summary of source row count, extracted row count, count differences, and failed prompt counts; null_or_empty_rows.md lists row indices where prompt extraction resulted in empty or null values. The data scale is 2011 extracted rows, with 0 failed prompt rows and 0 row count differences. This dataset is suitable for prompt engineering, reinforcement learning post-training, or language model fine-tuning tasks requiring multi-turn dialogue instruction following, particularly focusing on prompt extraction and structured representation. The dataset was generated by the Nemotron post-training v3 prompt extraction workflow and uploaded by jamesdborin.





